LECTURE II-11: SUPERSYMMETRIC FIELD THEORIES

Edward Witten

Notes by Pavel Etingof and David Kazhdan

11.1. General remarks on supersymmetry
Starting from today we study eld theories with supersymmetry, i.e. theories
whose symmetry group has a nontrivial extension to a supergroup. As usual, by
even and odd (innitesimal) symmetries we mean even, respectively odd, elements
of the Lie superalgebra of this supergroup.
It often happens that a solution to the classical eld equations has nontrivial
odd symmetries. Examples:
gradient owlines in Morse theory
holomorphic curves
instantons
monopoles
Seiberg-Witten solutions
hyperKahler structures
Calabi-Yau metrics
Metrics of G2 and Spin7 holonomy
In the next sections we will consider some of these examples.
11.2. Supersymmetric solitons (BPS states).
First consider a case in which the gradient owlines of Morse theory will appear.
Let h : Rn ! R be a Morse function (i.e. its critical points are nondegenerate and
jrhj grows at innity). A Morse function always has nitely many critical points.
Consider (in Minkowski signature) the theory of maps  : R2j2 ! Rn with the
Lagrangian
Z
(11.1) L = d2xd2 ( 21 D+D  h());
where D = @@  @ , @ = @x@ , x = 12 (t  x). This model has an obvious
supersymmetry under Q = @@ +  @ . These supersymmetry operators satisfy
the obvious commutation relations Q2 = @ , fQ+ ; Q g = 0.
We have (Q+  Q )2 = 2 @t@ . So if we look for classical solutions with time
translational symmetry (i.e. for solitons), we may in particular look for those of
them which are invariant under one of the supersymmetry, say Q+  Q .
We have
(11.2) @ + (   ) @ :
Q+  Q = ( @@  @@ ) + (+   ) @t + @x
+
Typeset by AMS-TEX
Thus, the supersymmetry condition for time-independent solutions is
(11.3) [( @  @ ) + (+   ) d ] = 0:
@ @
+ dx
Let us look for even solutions. It is easy to show that such solutions are of the form
 =  + + rh(). For them, the supersymmetry condition is

(11.4) d  rh() = 0;
dx
which is the condition for the gradient owline. Thus, supersymmetric solitons are
the owlines of the gradient ow.
Notice that these 1-st order equations imply the 2-nd order equations of motion.
Indeed, it is easy to show that the equations of motion are
(11.5) @+ @  + r(rh)2() = 0;
or for time-independent solutions

(11.6) d2 = r(rh)2();

dx2
which can be obtained by dierentiation of (11.4) with respect to x.
Another way to see this: the Lagrangian for time-independent elds (i.e the
Hamiltonian) is
Z
(11.7) H () = dx( 12 ( d
dx ) 2 + 1 (rh())2 ):
2
Rewriting H , we get
1 Z d Z1
(11.8) 2
H () = 2 dx( dx  r())  dh(s):
1
Since the last term is locally constant on the space of elds of nite energy, a
supersymmetric solution provides the global minimum for the energy in each con-
nectedR component of the space of elds. The value of energy at this minimum is
S = j dhj.
Denition. Supersymmetric solitons, that is classical solutions invariant under
some supersymmetries, are called classical BPS states.
11.3. The role of BPS states in quantum theory.
We have mentioned above and used the fact that the vector elds Q+ ; Q com-
mute. Since the space of solutions of the classical eld equations is a symplectic
supermanifold, these vector elds must be (at least locally) generated by some
Hamiltonian functions Q~ +; Q~ (dened up to adding a locally constant function).
But these functions need not Poisson commute: their Poisson bracket has to be a
locally constant function, not necessarily equal to zero.
2
 In our case, the functions Q~ +; Q~ are easy to write down: if  =  + + + +
 + + F , then
Z Z
(11.9) Q~+ = dx( + @+ + rh()); Q~ = dx( @  + rh()):
R
The computation of the Poisson bracket gives fQ~ +; Q~ g = 2S , S = dh =
h((1)) h(( 1)).
From this picture it is clear what will happen with operators Q+ ; Q in quantum
theory. We will have Q2+ = P+; Q2 = P (where P+; P are the corresponding
momentum operators), and (Q+  Q )2 = 2(H  S^), where H = (P+ + P )=2 is
the Hamiltonian, and S^ commutes with local operators.
In quantum theory, to every connected component Xa of the space X of elds of
nite energy there corresponds a summand Ha of the Hilbert space. We expect that,
if there is no breaking of supersymmetry, to every supersymmetric soliton  2 Xa
there corresponds a state in Ha which is also supersymmetric: (Q+  Q ) = 0.
Then (H  Sa) = 0, where Sa = S^jHa is a scalar. Therefore, since H  0, we
have H = jSaj . In particular, for every connected component of X there is only
one supersymmetry (out of the two), for which there can be supersymmetric states
in this component.
In general, on Ha we have H  jSaj.
Now we want to determine whether there is a supersymmetric quantum state,
that is a quantum state annihilated by Q+  Q , corresponding to the super-
symmetric classical state. In the lowest order of perturbation theory one nds no
bosonic zero mode except the translations and no fermionic zero mode. Hence,
in that approximation the lowest energy state of given momentum is unique and
nondegenerate. Also, the theory in the vacuum sector has a mass gap (classically
and therefore for weak enough coupling) so the unique ground state that is found
in the leading approximation is isolated from any continuum. Hence the quantum
theory for weak enough coupling has an isolated and unique ground state in this
sector of the Hilbert space, and it follows from the supersymmetry algebra that
this state must be annihilated by Q+  Q . This massive state is called a quantum
BPS state.
However, if (in a family of theories) two critical points of h collide and become
a degenerate critical point, the mass of BPS paths between them goes to zero, so
we can expect that the corresponding sector of Hilbert space loses its mass gap,
and massless particles appear. At such a point, supersymmetric states can appear
or disappear in the quantum theory. We will make this more explicit later in the
context of N = 2 supersymmetry.
11.4. N=2 supersymmety in 2 dimensions.
Consider the space R2j4 with coordinates x+ ; x ; +;  ; +;  . This space ad-
mits an action of the N=2 supersymmetry algebra, with supersymmetry generators

Q+ = @@ + + @x@ ; Q + = @@ + + @x@ ;

+ + + +
(11.10) @ @ @
Q = @ +  @x ; Q = @  +  @x ; @

3
Let us also introduce vector elds
D+ = @@ + @x@ ; D + = @ @ + @x@ ;
+ + + +
(11.11) @ @ @
D = @  @x ; D = @   @x ; @

which commute with the supersymmetry generators. Therefore, any Lagrangian

which is written in terms of the D's is supersymmetric.
Recall that a chiral function (or supereld) on R2j4 is a function satisfying the
equations D + = D  = 0. A general solution to these equations has the form
 =  ++ @+   @  + + +  @+ @  + +  F +
(11.12) + + +  ++  @+   + @ +:
You can read more about chiral functions in the superhomework.
Consider the theory of chiral maps  of R2j4 into C n , with the Lagrangian
1 Z Z Z
2 4  2 2   
(11.13) 2 d xd  + d xd+ d W () + d xd+d W ();
where W is a holomorphic function on C n , called the superpotential. This La-
grangian is N=2 supersymmetric. In components (for x+ = z; x = z), it looks
like
1 Z
(11.14) 2 d2x(jdj2 jF j2 + W 0 ()F + W 0 ()F + terms with fermions):
Setting all fermions to zero and the \dummy" eld F to the stationary point F =
W 0 (), we get the bosonic energy functional
1 Z
(11.15) H = 2 d2x(jdj2 + jW 0 ()j2):
It is easy to see that this functional coincides with (11.7), for the function h =
Re(e i W ), where  is any real number. Thus, at the classical level we are doing a
special case of the previous problem. However, now we have more supersymmetry
and therefore a more interesting theory.
11.5. N=2 BPS states
In the theory we are considering, there is an important symmetry called the
R-symmetry. It acts according to + ! ei + ;  ! e i  . If we require that
 is unchanged under this symmetry (i.e. Bose elds are unchanged and + !
e i + ; ! ei ), then Lagrangian (11.13) is obviously invariant under the
symmetry.
The commutation relations for the supersymmetry Hamiltonians are
fQ ; Q g = 2P;
fQ+; Q g = fQ +; Q g = 0 (by R-symmetry) ;
(11.16) fQ+; Q g = T; fQ +; Q g = T;

4
where T is analogous to S in Section 11.2 { it is a locally constant function on the
space of classical solutions (which is, unlike S , not necessarily real), and for brevity
we drop twiddles over Q's. Also, the squares of all the Q's are zero.
In fact, the function T is easy to compute, like the function S in the previous
problem. Namely,
(11.17) T = W ((1)) W (( 1)):
Now we will look at supersymmetric states. Choose a real number  and look
for states which are invariant under two supersymmetries Q1 () == Q+ + eiQ ,
Q2 () = Q + e i Q .
We have
(11.18) fQ1 ; Q2 g = 2i(H Re(e i T ));
which implies that supersymmetric classical states have to be time-independent.
The equation Q1 = 0 for an even function  gives (in the time-independent
case):

(11.19) @ = e i W 0 ();
@x
and the second equation gives the same result. This implies, in particular, that
Z Z1
(11.20) T = W ((1)) W (( 1)) = W 0 () d dx = e i
dx jW 0 ()j2dx;
1
which implies that  = argT . Thus, from equation (11.28) we get that for a
supersymmetric solution of the classical equations we have
(11.21) H = Re(e i T ) = jT j:
For other states in the connected component of this solution we have H  jT j.
Now, what are the supersymmetric solutions (of nite energy) geometrically? It
is clear from equation (11.19) that they are separatrices between critical points of
W for the gradient ow of h = Re(e i W ).
Now let us turn to quantum theory. From classical considerations we saw that
in our theory H  jT j, and in a nondegenerate case H = jT j only for BPS states.
Therefore, we should hope that in quantum theory the same situation takes place,
apriori with a corrected value of T .
Consider the generic situation when all zero-modes of the Hamiltonian near a
classical BPS state arise from the superPoincare group. This is the case if for any
3 critical points a; b; c of the potential we have jTacj < jTabj + jTbcj, where Tab is
the value of T on the component of the space of solutions which go from a to b. In
other words, this is the case when the gradient owline between a; b never passes
through c.
In this case, in quantum theory, for small values of the coupling, we expect
that the point jT j in the spectrum of H occurs discretely. From our classical
5
computations, we expect that the eigenspace corresponding to this eigenvalue is
nite-dimensional, and is an irreducible representation of the odd part of the su-
perPoincare algebra, in which Q1 ; Q2 act by zero. This representation is nothing
but the space of sections of the equivariant vector bundle on the upper part of
the hyperboloid, where the ber is the standard 2-dimensional irreducible repre-
sentation of the 4-dimensional Cliord algebra generated by the two remaining
supersymmetries.
Notice that all other superPoincare representations occuring in this theory have
to be not 2-dimensional but rather 4-dimensional over the ring of functions on the
hyperboloid, since for levels of energy above T the Cliord algebra satised by the
supersymmetry operators corresponds to a nondegenerate quadratic form, and the
only irreducible represenation of this algebra is 4-dimensional.
11.6. N=1 Supersymmetry in 4 dimensions.
Now consider supersymmetry in 4 dimensions. We start with N = 1 supersym-
metry. In this case the odd part of the supersymmetry algebra is similar to the
N = 2 case in two dimensions. It is generated by Q , Q , with relations
fQ; Q g = 2P ;
fQ ; Q g = 0;
(11.22) fQ  ; Q  g = 0:
where ;  2 f+; g, and P is a basis of the space of complex linear functions on
the spacetime (These relations exhibit the isomorphism of Poincare representations
S+
S ! VC , where S are the spinor representations.) A central extension like
in (11.16) cannot arise here because it is prohibited by the Poincare symmetry (this
central extension would have to be in the representation S 2C 2 of SU (2) (where C 2
is the standard representation), which is the spin 1 representation and contains no
invariants).
The determinant of the quadratic form corresponding to the Cliord algebra
(11.22) is (P 2 )2 . The rank of this form if P 2 = 0 is 2. Therefore, massless repre-
sentations of the superPoincare correspond to 2-dimensional representations of the
Cliord algebra, and massive representations correspond to 4-dimensional repre-
sentations of the Cliord algebra.
More precisely, consider a representation W of the superPoincare and the sub-
space Wp on which P = p, where p 2 R1;3. Let Gp be the stabilizer of p in the
group of rotations, Hp the maximal torus in Gp (always isomorphic to U (1)). With
respect to Hp , W has a decomposition in a direct sum of representations of integer
and half-integer spins. These spins are called helicities of W , and each helicity has
a multiplicity.
It is easy to see that the supersymmetry operators Q ; Q  raise helicity by 1=2
(since they live in the spinor representation of the Poincare). Thus, the mass-
less representations have helicities j; j + 1=2 with multiplicity 1, and the massive
representations have helicities j; j + 1=2; j + 1 with multiplicities 1; 2; 1.
Now let us consider massless particle multiplets which are allowed by N = 1
supersymmetry. There are two such basic multiplets.
1. Vector multiplet: a gauge eld A and a chiral spinor  in the adjoint repre-
sentation. This is the N=1 analogue of the gauge eld. In particular, in the U (1)
6
case the theory is free. In the infrared, it generates a massless vector and a mass-
less spinor, so the helicities are 1; 1 (for vector) and 1=2; 1=2 (for spinor). In
particular, the massless representation of the SuperPoincare arising in this theory
is reducible, and splits into two: the one with helicities 1; 1=2 and the one with
helicities 1=2; 1. However, over R this splitting does not exist and our represen-
tation is irreducible. A more physical version of this statement is to say that the
helicity 1; 1=2 states are related to helicity 1=2; 1 states by CPT conjugation, so
that one pair must be present if the other is. Thus, this eld conguration is the
minimal supersymmetric one which contains a gauge boson.
2. Chiral multiplet: A massless complex scalar  and a massless chiral spinor
. In this case the story is analogous. The obtained massless representation has a
spinor and a scalar, so it has helicities 0; 0 (for scalar) and 1=2; 1=2 (for spinor).
Thus this representation is again a sum of two, with helicities 1=2; 0 and 0; 1=2.
This decomposition is only valid over C and not over R; over R, the representation
is irreducible, so this combination is the smallest supersymmetric one containing a
scalar.
The massive versions of these multiplets are as follows.
1. Massive vector multiplet (or hypermultiplet). The minimal real supersym-
metric representation containing a massive vector has helicities 1; 1=2; 0; 1=2; 1
with multiplicities 1; 2; 2; 2; 1. This involves a massive vector, two massive spinors
(chiral and antichiral), and a real massive scalar. The corresponding 8-dimensional
representation of the Cliord algebra is (over C ) a sum of two 4-dimensional rep-
resentations.
2. Massive chiral multiplet. Same as massless chiral multiplet (we could consider
the same elds with masses, such that the mass of bosons equals the mass of
fermions).
11.7. N=2 Supersymmetry in 4 dimensions.
In the N=2 case, we have two copies of operators Q: Q(1) and Q(2), and they
commute in the same way as before if the indices are equal, and give zero commu-
tator if they are not equal (as vector elds). However, now there is a possibility for
a central extension: it is no longer prohibited by the Poincare symmetry.
Consider rst the case when the central term is zero. In this case the quadratic
form of the Cliord algebra is nondegenerate (in 8 dimensions) in the massive case,
and has rank 4 in the massless case. Thus, an irreducible massive representation
should be 16-dimensional, and an irreducible massless representation should be 4-
dimensional. The helicities for representations are obained like in N = 1 case.
In particular, in a massive representation the helicities are in groups of the form
(j,j+1/2,j+1,j+3/2,j+2) with multiplicities (1,4,6,4,1), and in a massless represen-
tation they are in groups (j,j+1/2,j+1) with multiplicities (1,2,1).
Now let us consider the simplest N = 2 supersymmetric theory. It is obtained by
combining elds from a (classically) massless vector multiplet and a massless chiral
multiplet in adjoint representation. Thus the elds are: from vector multiplet {
(A; ), from chiral multiplet { (; ). The Lagrangian is the minimal N = 2 su-
persymmetric Lagrangian on these elds. Such a Lagrangian exists and is uniquely
determined by the minimality condition. In the U (1) case the theory is free, but in
the nonabelian case there are nontrivial interactions.
What does this theory do in the infrared? In the U (1) case, the answer is simple:
7
we should add together the representations for the vector and chiral multiplets. We
get helicities (-1,-1/2,0,1/2,1) with multiplicities (1,2,2,2,1). This is the sum of
two complex conjugate massless representations of the N=2 Cliord algebra: the
one with helicities (0,1/2,1) and multiplicities (1,2,1) and the one with helicities
(-1,-1/2,0) and multiplicities (1,2,1).
But now let us consider the nonabelian case (say the gauge group G is simple).
Then the bosonic elds are the gauge eld A and a scalar  in the complexied
adjoint representation gC . The bosonic part of the Lagrangian is
Z 1
(11.23) Lbosonic = ( 4e2 F 2 + jdA j2 + j[; ]j2):
Thus the bosonic part of the space of classical vacua is the set of solutions of
the equations [; ] = 0 modulo the action of G. This space can be identied with
tC =W , where t is the maximal commutative subalgebra in g and W the Weyl group.
In the case G = SU (2), this quotient is identied with C by introducing the global
coordinate u = Tr(2 ) (the u-plane).
Recall from Lecture 2 that in this situation we have gauge symmetry break-
ing from G to the centralizer H of . In particular, for SU (2) near u 6= 0 the
gauge symmetry is broken classically from SU (2) to U (1) (Higgs mechanism). The
components of the gauge eld which are charged nontrivially with respect to the
surviving U (1) will become massive. By N=2 supersymmetry, the same will happen
for the corresponding components of ; ; . Thus, in the charge 2 and 2 sectors
of the Hilbert space the lowest energy states will be in a massive representation
with helicities as before: (-1,-1/2,0,1/2,1) with multiplicities (1,2,2,2,1). However,
we know that there is no such representation without central extension. Thus,
without central extension we get a contradiction, and hence the central extension
must appear.
The central extension will show up in the commutation relations between Q(1) 
(2)
and Q :
fQ(1) (2)
(11.24)  ; Q g = " Y;
where Y is an operator which commutes with all local operators (central charge).
Before the central charge, the algebra had a U (2) R-symmetry (action on indices
1 and 2), but the central charge breaks this symmetry down to SU (2) (the chiral
U (1) symmetry is anomalous in our theory, because of the index problems, and it
is broken to Z=2Z).
It is easy to see that in this case the massive particles considered above must
have mass exactly jY j (this is where the quadratic form of the Cliord algebra
becomes degenerate).
It turns out that classically there are BPS states which correspond to these
massive particles. Namely if we are at the vacuum u 2 C , we have
 
(11.25) 1
 p 0 a : a 0
2
where a = u1=2. Let a = ei. We should look for BPS states (i.e. states which
are time-independent and invariant under half of the supersymmetry) which satisfy
8
the condition (11.25) at innity. If Y = jY jei then the equation of being invariant
under half of the supersymmetry is
(11.26) F = e i  dA;
where  is in R3. (the BPS monopole equations). Since F is real, we must have
 =  modulo .
It can be shown that such BPS states exist in sectors with charges 2 and 2
(charges are the eigenvalues of the innitesimal operator corresponding to the un-
broken U (1)-gauge symmetry). When these solutions are quantized, we will get the
same result as in 2 dimensions. Namely, in quantum theory, we will get a repre-
sentation of the superPoincare with helicities coming in groups (j,j+1/2,j+1) with
multiplicities (1,2,1). Adding two copies of such groups with j = 1 and 0, we
will get the massive hypermultiplet (=massive vector multiplet); this multiplet has
the right helicities, which we found by considering the Higgs mechanism. Thus, in
presence of central charge we get no contradiction.
Remark. Computing the commutators of the Q-s using currents, one can show
that classically
Z
(11.27) Y = (  F + e12 F );

where  is a distant sphere in the space cycle. So it is a combination of the electric
and magnetic charge for the eective U (1) theory.